Anal. Chem. 1986, 58,2855-2858 (7) Yatslmirskii, K. B. Kinetic Methods of Analysis; Pergammon: Oxford, 1966. (8) First International Symposium on Kinetics in Analytical Chemistry; Cordoba, Spain, September 27-30, 1983. (9) Carr, P. W. Anal. Chem. 1078, 50, 1602-1607. (10) Blaedei, W. J.; Olson, C. Anal. Chem. 1963. 36, 343-347. (11) James, G. E.; Pardue, H. L. Anal. Chem. 1968, 40, 796-802. (12) Crouch, S. R. Anal. Chem. 1969, 4 7 , 880-883. (13) Parker, R. A,; Pardue, H. L.; Willis, B. G. Anal. Chem. 1970, 42, 58-61. (14) Cordos, E. M.; Crouch, S. R.; Maimstadt; H. V. Anal. Chem. 1868, 40, 1812-1818. (15) Ingle, J. D., Jr.; Crouch, S. R. Anal. Chem. 1972, 42, 1055-1060. (16) Malmstadt. H. V.; Crouch, S. R. J. Chem. Educ. 1966, 43, 340-353. (17) Calicott, R. H.: Carr, P. W. Anal. Chem. 1974, 46, 1840-1842. (18) Iracki, E. S.; Malmstadt, H. V. Anal. Chem. 1973, 45, 1766-1770. (19) . . Atwood. J. G.; DiCesare, J. L. Clin. Chem. (Winston-Salem, N.C.) 1973, 19, 1263-1269. (20) Landis, J. B.; Rebec, M.; Pardue, H. L. Anal. Chem. 1977, 49, 785-788. (21) Holier, F. J.; Calhoun, R. K.; McClanahan, S. F. Anal. Chem. 1982, 54. 755-761. (22) Davis, J. E.: Renoe, B. Anal. Chem. 1979, 57,526-528. (23) Mieiing, G. E.; Pardue, H. L. Anal. Chem. 1978, 50, 1611-1618. (24) Wentzell, P. D.; Crouch, S. R. Anal. Chem., following paper in this issue. (25) Ingle, J. D., Jr.; Crouch, S. R. Anal. Chem. 1971, 43, 697-701. (26) Pausch, J. 8.; Margerum, D. W. Anal. Chem. 1989, 4 7 , 226-232.
2855
(27) Rieman, W.; Beukenkamp, J. I n Treatise on Analyfical Chemishy; Kokhoff, I.M., Eking, P. J., Eds.; Wiley: New York, 1961; Part 11, Vol. 5, pp 346-351. (28) Javier, A. C.; Crouch, S. R.; Malmstadt, H. V. Anal. Chem. 1969, 4 7 , 239-243. (29) Beckwith, P. M.; Crouch, S. R. Anal. Chem. 1972, 44, 221-227. (30) Balciunas, R. Ph.D. Dissertation. Michigan State University, East Lansing, MI, 1981. (31) Putt, R. M.S. Thesis, Michigan State University, East Lansing, MI, 1983. (32) Newcome, B.; Enke, C. G. Rev. Sci. Insfrum. 1984, 55, 2017-2022. (33) Nevius, T. A.: Pardue, H. L. Anal. Chem. 1984, 56, 2251-2253. (34) Neider, J. A.; Mead, R. Computer J. 1985, 7, 308-313. (35) Beckwith, P. M.; Sheeline, A.; Crouch, S. R . Anal. Chem. 1975. 47, 1930- 1936. (36) Kircher, C. C.; Crouch, S. R . Anal. Chem. 1983, 55, 242-248.
RECEIVED for review January 21,1986. Accepted July 7,1986. The authors gratefully acknowledge the financial support of the National Science Foundation through NSF Grant No. CHE 8320620 and the Natural Sciences and Engineering Research Council of Canada through an NSERC Graduate Fellowship (P.D. W.).
Comparison of Reaction-Rate Methods of Analysis for Systems Following First-Order Kinetics Peter D. Wentzell and S. R. Crouch* Department of Chemistry, Michigan State University, East Lansing, Michigan 48824
Several measurement and computational approaches to reactlon-rate methods of chemlcal analysis are crttically compared for reactionsfoliowlng firstorder ktnetics. Both traditional technlques (fixed-time, variable-time, inltlal rate, and derivative methods) and more recent methods, which minlmize dependence on rate constant variations, are considered. The theoretical and experimental performance of each method Is evaluated under condttions of betweerrrun variations in the rate constant and wlth an invariant rate constant. Addltlonal factors that Influence the choke of a reaction-rate method for a particular analytical application are also dlscussed. of the seven methods examined, the Cornell method of partial sums exhlblted the best overall performance in the areas investigated.
The popularity of reaction-rate methods in chemical analysis is indicated by the wide variety of methodologies that have been developed over the years (1-6). Among the most widely used rate methods are the fixed-time (1, 7, 8), the variable-time ( I , 9-11), and the derivative ( I , 12,13) methods. These traditional reaction-rate methods are often susceptible to variations in experimental parameters that affect the rate constant, such as pH, temperature, ionic strength, and reagent concentration. Carr (7) has shown how small variations in the first two variables can adversely affect the precision of the results. In recent years, several workers have attempted to alleviate the problem of between-run variations in the rate constant for systems following first- or pseudo-first-order kinetics. Atwood and DiCesare (14) first noted that the effect of rate constant variations could be minimized by appropriate adjustment of enzyme activity in substrate determinations by kinetic methods. They suggested that the enzyme activity 0003-2700/8610358-2655$01 SO10
should be adjusted so that the reciprocal of the pseudofirst-order rate constant is equal to the time of measurement. Pardue and co-workers (15)extended this observation, noting that for any reaction following first-order kinetics, the optimum rate measurement time for minimizing the effect of between-run variations in the rate constant is t = l / k = T . This observation was investigated further by Holler et al. (16), who demonstrated experimentally that improved results could be obtained with the method. For convenience, this method of measuring rates at t = T is referred to as the optimized derivative method throughout this paper. Davis and Renoe (17) described an optimized fixed-time approach which reduces the influence of between-run rate constant variations for the traditional fixed-time method. This method optimizes measurement times for minimal rate constant dependence. Mieling and Pardue (18) have proposed a multiple-linearregression procedure which evaluates kinetic parameters to compensate for changes which occur in the reaction curve. A method developed by Cornell (19) in 1962 for fitting exponentials through partial sums is also applicable to the problem of between-run variations in the rate constant. While the Cornell method has been extended to other kinetic systems by Kelter and Carr (20-22), it has not appeared extensively in the analytical chemistry literature and in this sense its application to reaction-rate methods of analysis is fairly recent. Wentzell and Crouch (23) have introduced a new method, the two-rate method, which attempts to eliminate the dependence on between-run variations in the rate constant by using rate measurements made a t two times during the course of the reaction. In this paper, we provide a critical comparison of several reaction-rate methods applied to systems following first- or pseudo-first-order kinetics. Both traditional methods (fixed-time, variable-time, derivative, and initial rate methods) and the more recently developed techniques (optimized de@ 1986 American Chemical Society
2858
ANALYTICAL CHEMISTRY, VOL. 58, NO. 13, NOVEMBER 1986
rivative, optimized fiied-time, Cornell, and two-rate methods) are considered. Some of the more specialized methods, such as the kinetic difference (24, 25) and signal-stat (26, 27) methods, are not included in this comparison. The regression-kinetic method of Mieling and Pardue (18) is also excluded, since its level of sophistication is beyond the scope of this treatment. Otherwise, we have attempted to be as general as possible, evaluating each method in terms of susceptibility to rate constant variations, precision in the absence of those variations, and several other factors. The experimental results presented are limited to spectrophotometric data, but many of the conclusions drawn are; generally applicable. Throughout this discussion, reaction-rate methods are classified in two ways for ease of reference. First, adopting the terminology of Pardue et al. (B), we define those methods requiring an instantaneous measurement of reaction rate (derivative, optimized derivative, initial rate, and two-rate methods) as differential methods, while the others (fixed-time, variable-time, optimized fiied-time, and Cornell methods) are referred to as integral methods. In addition, “traditional” methods (fiied time, variable time, initial rate, and derivative) are distinguished from “optimized” methods (optimized derivative, optimized fixed time, Cornell, and two rate), which attempt to minimize susceptibility to rate constant variations.
EXPERIMENTAL SECTION Two chemical systems were used to evaluate the reaction-rate methods discussed in this paper. Phosphate determinations were carried out by means of the 12-molybdophosphate reaction (29, 30). Calcium determinations used the metal/complex exchange reaction described by Pausch and Margerum (31). All reactions were carried out on an automated stopped-flow system (32,33) and data were processed on a DEC LSI 11/23 computer. Further experimental details have appeared elsewhere (23). RESULTS AND DISCUSSION Errors in a reaction-rate method can be divided into two broad categories: (1)errors arising from between-run variations in the rate constant, and (2) errors arising from other sources. Each of these is considered for the methods compared. In addition, other factors that may influence the selection of a rate method are discussed. Susceptibility to Rate Constant Variations, A theoretical comparison of the effect of between-run constant variations on seven different reaction-rate methods was carried out to evaluate the extent of the dependence in each case. The methods examined were the fixed-time, variable-time, initial-rate, optimized derivative, optimized fixed-time, Cornell, and two-rate methods. Figure 1 compares the theoretical percent error in the concentration determined by each method as a function of the deviation of the rate constant from its nominal value, ko. To generate this plot, measurement times of tl = 0 and tz = l / k o were assumed for the fixed-time method. Also, for the variable-time method, the reaction was assumed to be 2% complete at the time of the measurement. Finally, for the optimized fixed-time method, measurement times of 0.5/ko and 1.76/k0, values given by Davis and Renoe (17), were chosen. Examination of Figure 1leads to some interesting conclusions. Of the six methods, the variable-time method and the initial rate method exhibit the largest dependence on the rate constant. For these methods, the reaction curve is very nearly linear in the measurement region and, therefore, the measured parameters are directly proportional to the first-order rate constant. The fixed-time method shows a somewhat smaller dependence on k , but this is highly dependent on the selection of measurement time. Short measurement intervals increase the dependence on the rate constant, approaching that of the variable-time method a t the limit. As the interval becomes
:
100
--
0 C 0
0
c ._
I L
0 -
?
W
O
R
-1
Y
1
1
c
0.0
’ .o
I 2 0
Flgure 1. Theoretical errors expected in the concentration determined by various reaction-rate methods as a function of of the rate constant from its nominal value.
the relative deviation
longer, there is an approach to equilibrium methods, which have virtually no dependence on the rate constant. The reason for improved results with the optimized derivative and optimized fixed-time methods is easily seen from Figure 1. Both of these methods exhibit an asymptotic approach to zero error at the point of zero deviation in the rate constant. Because of the flatness of the curves in this region, the effect of variations in the rate constant is minimized, although it is not zero. The similarity of the two curves is worth noting since it indicates an approximately equivalent dependence on k for the two methods. The optimized fixed-time method is expected to improve in this respect as a wider measurement interval is chosen, but again, this is due to an approach to equilibrium methods. Another point worth noting with regard to these two methods is that negative errors in the concentration result regardless of the direction of the variation of the rate constant. This is an undesirable feature since it introduces bias. The Cornell and two-rate methods show, in theory, no errors attributable to between-run variations in the rate constant. The two-rate method compensates for these variations by making rate measurements at two times during the reaction and is limited in practice by random errors in the rate measurements (23). The Cornell method adjusts to between-run changes in the rate constant by employing a relatively simple algorithm to fit the exponential data. The practical limitations of this method have not been extensively investigated. Table I illustrates experimentally the performance of several reaction-rate methods under conditions of gross changes in the rate constant. The results were obtained using the molybdophosphate reaction for the determination of phosphate. The rate constant was varied by changing the temperature. Calibration curves were generated for each method a t a temperature of 26.0 “C. “Unknowns” of 4 Mg/mL and 8 Fg/mL P were then analyzed a t this temperature and also at 21.5 OC and 31.2 OC. The errors in the measured phosphate concentrations are reported in Table I. As anticipated, the variable-time and initial rate methods exhibit the greatest errors when the temperature is changed. As predicted by Ingle and Crouch (34),the variable-time method also gives poor results at the temperature at which the calibration curve was generated, which demonstrates that this method is best applied to systems following zero-order kinetics. The fixed-time results (at t = T) are somewhat better than the variable-time results, but the errors are still large. The errors resulting from the application of the optimized derivative ( t = T) and the optimized fixed-time ( t l = 0.57,t z = 1.767) approaches are, as predicted, comparable and more reasonable. Also as pre-
ANALYTICAL CHEMISTRY, VOL. 58, NO. 13, NOVEMBER 1986
Table I. Comparison of Results Obtained by Various Reaction-Rate Methods under Conditions of Large Changes in Temperature, Based on Calibration Curves Generated at 26.0 OC
-
-
% error in uhosuhate measured
method variable time initial rate fixed time (17) fixed time (2T) optimized fixed time (0.57,1.767) optimized derivative two-rate method (0.757,1.55) Cornell method (range = 2.787)
phosphate T = 21.5 taken, @g/mL "C
T = 26.0 5" = 31.2 OC
OC
4.00 8.00 4.00 8.00 4.00 8.00 4.00 8.00 4.00 8.00
-43 -45 -49 -49 -40 -33 -24 -22 -14 -16
-4.1 -7.8 2.7 0.5 0.4 0.4 -0.5 -0.2 -3.0 1.5
59 52 78 88 42 39 16 18 -6.2 -12
4.00 8.00 4.00 8.00
-14 -14 -1.6 -0.6
1.1 -0.1 -1.5 -1.7
-12 -12 6.3 7.2
4.00 8.00
-0.1 -0.4
0.6 0.16
6.0 5.1
dicted, there is a consistently negative bias with these two methods. Finally, the two-rate method (tl = 0.757, t2 = 1.57) and the Cornell method (range of data used = 2.787) show the least dependence on changes in the rate constant, although the errors at the high temperature seem inordinately large. These high-temperature deviations have been previously discussed for the two-rate method (23). For the Cornell method, the errors are believed to be due to a limitation of the algorithm which becomes more important as the rate constant increases. Precision with Invariant Rate Constant. We now consider the case where between-run variations in the rate constant are negligible and errors arising from other sources, such as detector noise or photometric source fluctuations, limit precision (28,s To )examine . precision in the absence of rate constant variations, ten replicate sets of data were obtained for the calcium determination. Rate constant fluctuations were minimized by maintaining careful pH and temperature control. The data obtaiped were analyzed by several methods, and the relative standard deviation (RSD) calculated in each case is reported in Table 11. Among the differential methods, four techniques are compared: the derivative method (at t = 0 . 5 ~ )the , optimized derivative method, and the two-rate method with two sets of measurement times. The initial rate method is not included since the method of rate calculation is significantly different and would not allow a reliable comparison. Rates for the methods compared were computed by employing a digital version of the up/down integration method (35, 36), which is equivalent to a modified Savitzky-Golay derivative filter (37,38). For our system, we found that the absolute errors in the rate measurements were largely independent of the time of measurement. This is a consequence of the method of rate calculation and the fact that absolute errors in the absorbance measurements remained effectively constant over the small absorbance range used. In Table 11,the two-rate method shows similar or improved precision over the optimized derivative method, as predicted elsewhere (23). A single rate measurement at an earlier time shows improved precision over both of these due to a larger rate with comparable errors in the rate measurements. However, this advantage is overshadowed by a greater susceptibility to rate constant variations.
2857
Table 11. Precision of Various Reaction-Rate Methods under Conditions of an Invariant Rate Constant method
parameters
70 RSD
derivative optimized derivative two rate two rate fixed time fixed time optimized fixed time optimized fixed time OFT (smoothed) Cornell Cornell Cornell
t = 0.57 t=T tl = 0.55,tz = IT ti = 0.55,tz = 1.767 ti = 0, tz = 17 ti = 0, tz = 27 ti = 0.87,tz = 1.237 ti = 0.55,ti = 1.767 ti = 0.57,tz = 1.76~ range = 2.257,150 points range = 4.57,300 points , points range = 6 ~399
0.57 1.1 1.1 0.86 1.2 0.83 5.0 1.7 0.41 0.34 0.38 0.67
Among the integral methods considered, Ingle and Crouch (34)have already demonstrated the inferior precision of the variable-time method when applied to first-order systems. In the current study, the errors would be compounded by a relatively long sampling interval (poor time resolution), so the results for this method are not included in Table 11. The remaining integral methods exhibit a similar compromise between precision and rate constant dependence as was seen for the differential methods. Pardue et al. (28) have shown that for small absorbance changes, as is the case here, relative errors can be expected to be inversely proportional to the magnitude of the absorbance change. This is supported by the results in Table 11. Both the fixed-time and optimized fixed-time methods show improved precision when the time interval is increased, approaching the precision of equilibrium methods in the limit. Also, since the fixed-time method (with tl = 0) utilizes larger absorbance changes than the optimized fixed-time method for equivalent time intervals, the latter method generally exhibits poorer precision in the absence of rate constant variations. The Cornell method exhibits the best precision of the methods examined, primarily because all of the data are used. The precision obtained will be a function of the number of points and the range of data employed in determining the partial s u m , as shown in the table. The exact nature of this dependence warrants further investigation. A direct comparison of the precision observed for differential and integral methods would be unfair since the method of parameter calculation is not the same. Differential methods require differentiation of the analytical signal to compute the instantaneous rate. Since this process amplifies high-frequency noise (6,38),some smoothing of the data is normally required. In this work, smoothing was carried out concurrently with differentiation by a modified Savitzky-Golay derivative filter (37, 38). To obtain a fair comparison of the precision between differential and integral methods, it was necessary to apply a comparable smoothing filter to the data for the latter. The last entry for the optimized fixed-time method in Table I1 gives the relative standard deviation of the results when such a filter was applied. The filter used was a simple Savitzky-Golay (39)linear smooth. The result indicates that, when compared on common ground, the integral methods are more precise than the differential methods. This observation is expected to be generally valid due to the lower precision of instantaneous rate measurements. Additional Factors. Several other factors may be of importance when consideringwhich rate method to use in a given application. For example, the compatibility of the rate computation technique to continuous flow methods of analysis may be important if determinations are to be carried out in flowing streams. Of the methods discussed here, only the fixed-time method and the optimized fixed-time method are directly compatible with flow injection analysis (FIA) (40) and air-segmented continuous flow analysis ( 4 1 ) . The other
2858
ANALYTICAL CHEMISTRY, VOL. 58, NO. 13, NOVEMBER 1986
methods can be implemented in flow systems only if the flow is stopped. Another important factor is the instrumental and computational complexity of the rate methods. All of the methods discussed have the same fundamental components in the instrumental design, so the complexity involved in their implementation is largely determined by the amount of preand/or postprocessing of the data required. The integral methods are all implemented with relative ease. With the exception of the Cornell method, each requires a simple calculation of the difference between two measurements. Differential methods, on the other hand, require differentiation of the analytical signal. Many effective electronic differentiation methods have been developed (35,36,42-44), but all require careful implementation to avoid degradation of precision due to noise amplification. Numerical methods (37-39) are another option, but again careful selection of the digital filter is required to ensure sufficient noise immunity and minimal signal distortion. Thus, the necessity for a direct rate calculation is a complicating aspect of differential methods.
CONCLUSION Each of the reaction-rate methods examined exhibits its own particular advantages and disadvantages when applied to systems following first-order kinetics. An exception to this is the variable-time method, which is more applicable to zero-order reactions. As expected, the optimized methods are less sensitive to between-run fluctuations in the rate constant than traditional methods but generally show poorer precision in the absence of these fluctuations. In this study, the Cornell method exhibited the best overall performance. It is a multipoint method that is insensitive to between-run variations in the rate constant, gives precise results in the absence of those variations, and is relatively simple to implement. The analytical limitations of this method require further investigation, however. LITERATURE CITED Malmstadt, H. V.; Delaney, C. J.; Cordos, E. A. CRC Crit. Rev. Anal. Chem. 1972, 2 , 559-619. Pardue, H. L. Clin. Chem. (Winston-Salem, N.C.) 1977, 2 3 , 2189-2201. Carr, P. W.; Bowers, L. D. Immobilized Enzymes in Analytical and Clinical Chemistry; WHey-Interscience: New York, 1980; Chapter 3. Mark, H. B., Jr.; Rechnitz, G. A. Kinetics in Analytical Chemistry; Wiley-Interscience: New York, 1968. Pardue, H. L. I n Advances in Analytical Chemistry and Instrumentation; Reilley, C. N.; McLafferty, F. W.. Eds.; Wiley-Interscience: New York, 1964; Vol. 7, pp 126-140.
(6) Crouch, S.R. I n Computers in Chemistry and Instrumentation; Manson, H. D., Mark, H. B., Jr., MacDonakl. H. C.. Eds.; Marcel Dekker: New York, 1973; Vol. 3, pp 107-207. (7) Carr, P. W. Anal. Chem. 1978, 5 0 , 1602-1607. (8) Biaedel, W. J.; Olson,C. Anal. Chem. 1983, 36, 343-347. (9) James, G. E.; Pardue, H. L. Anal. Chem. 1968, 4 0 , 796-802. (10) Crouch, S. R. Anal. Chem. 1969, 4 1 , 880-883. (11) Parker, R. A.; Pardue, H. L.; Williams, B. G. Anal. Chem. 1970, 42, 56-6 1. (12) Pardue, H. L. Anal. Chem. 1984, 3 6 , 633-636. (13) Pardue, H. L. Anal. Chem. 1964, 3 6 , 1110-1112. (14) Atwood, J. G.;DiCesare, J. L. Clin. Chem. (Winston-Salem, N.C.) 1973, 19, 1263-1269. (15) Landis, J. B.; Rebec, M.; Pardue, H. L. Anal. Chem. 1977, 4 9 , 785-788. (16) Holier, F. J.; Calhoun, R. K.; McClanahan, S. F. Anal. Chem. 1982, 5 4 . 755-761. (17) Davis, J. E.; Renoe, B. Anal. Chem. 1979, 5 1 , 526-528. (18) Mieling, G. E.; Pardue, H. L. Anal. Chem. 1978, 50, 1611-1618. (19) Cornell, R. G. Biometrics 1962, 18, 104-113. (20) Keiter, P. B.; Carr, J. D. Anal. Chem. 1979, 5 1 , 1825-1828. (21) Kelter, P. B.; Carr, J. D. Anal. Chem. 1979, 5 7 , 1828-1834. (22) Keiter, P. B.; Carr, J. D. Anal. Chem. 1980, 5 2 , 1552. (23) Wentzell, P. D.; Crouch, S. R. Anal. Chem., preceding paper in this issue. (24) Weisz. H.; Ludwig, H. Anal. Chim. Acta 1971, 55, 303-313. 125) Weisz. H.: Pantel. S. Anal. Chim. Acta 1974. 68. 311-316. Malmstadt, H. V.: Piepmeier, E. H. Anal. Chem.' 1965, 37, 34-44. Weisz, H.; Rothmaier, K. Anal. Chim. Acta 1975, 7 5 , 119-126. Pardue, H. L.; Hewin, 1.E.; Milano, M. J. Clin. Chem. ( Winston-Sa/em, N.C.) 1974, 2 0 , 1028-1042. Rieman, W.; Beukenkamp, J. I n Treatise on Analytical Chemistry; Koithoff, I.M., Elving, P. J., Eds.; Wiiey: New York, 1961; Part 11, Vol. 5, pp 348-351. Javier, A. C.; Crouch, S.R.; Malmstadt, H. V. Anal. Chem. 1969, 41, 239-243. Pausch, J. B.; Margerum, D. W. Anal. Chem. 1969, 41, 226-232. Beckwith, P. M.; Crouch, S. R. Anal. Chem. 1972, 44, 221-227. Crouch. S. R.; Holler, F. J.; Notz. P. K.; Beckwith. P. M. Appl. Spectrosc. Rev. 1977, 13, 165-259. Ingle. J. D., Jr.; Crouch, S. R. Anal. Chem. 1971, 4 3 , 697-701. Cordos, E. M.; Crouch, S. R.; Maimstadt, H. V. Anal. Chem. 1968, 40, 1812-1818. Ingle, J. D., Jr.; Crouch, S. R. Anal. Chem. 1972, 42, 1055-1060. Nevius, T. A.; Pardue, H. L. Anal. Chem. 1984, 5 6 , 2249-2251. Hamming, R. W. Digltal Filters; Prentice-Hail: Englewood Cliffs. NJ, 1983. Savitzky, A.; Golay, M. J. E. Anal. Chem. 1964, 36, 1627-1639. Ruzicka, J.; Hansen, E. H. Flow Injection Analysis; Wiiey-Interscience: New York, 1981. Furman, W. B. Continuous Flow Analysis; Marcel Dekker: New York. 1976. Maimstadt, H. V.; Crouch, S.R. J . Chem. Educ. 1988, 43, 340-353. Caiicott, R. H.; Carr, P. W. Anal. Chem. 1974, 4 6 , 1840-1842. Iracki, E. S.; Maimstadt, H. V. Anal. Chem. 1973, 45, 1766-1770.
RECEIVED for review January 21,1986. Accepted July 7,1986. The authors gratefully acknowledge the financial support of the National Science Foundation through NSF Grant No. CHE 8320620 and the Natural Sciences and Engineering Research Council of Canada through an NSERC Graduate Fellowship (P.D. W.).